UBarney commented on PR #16889: URL: https://github.com/apache/datafusion/pull/16889#issuecomment-3121216474
> > I'll need some time to read through the code, but reduced memory usage is also impressive 👍 > > > > * v1 (old) > > > > ``` > > Query Time (ms) Peak RSS Peak Commit Page Faults > > -------------------------------------------------------------- > > 1 657.40 1112.0 MB 10.6 GB 0 > > 2 833.66 1691.1 MB 10.6 GB 0 > > 3 1688.13 3.4 GB 10.8 GB 0 > > 4 4176.38 16.2 GB 20.1 GB 0 > > 5 1629.22 23.2 GB 57.2 GB 0 > > 6 17687.00 18.9 GB 26.1 GB 0 > > 7 1245.64 23.3 GB 39.2 GB 0 > > 8 19420.73 20.1 GB 31.3 GB 0 > > 9 3467.28 10.4 GB 19.9 GB 0 > > 10 5414.65 44.7 GB 46.0 GB 0 > > ``` > > > > > > > > > > > > > > > > > > > > > > > > > > * v2 (nlj-rewrite-tmp) > > > > ``` > > Query Time (ms) Peak RSS Peak Commit Page Faults > > -------------------------------------------------------------- > > 1 383.61 104.4 MB 5.0 GB 0 > > 2 645.60 471.2 MB 8.0 GB 0 > > 3 763.10 1738.5 MB 6.0 GB 0 > > 4 1973.21 5.9 GB 9.6 GB 0 > > 5 644.86 145.1 MB 4.0 GB 0 > > 6 6043.58 149.3 MB 2.0 GB 0 > > 7 543.70 154.5 MB 7.0 GB 0 > > 8 6516.91 177.4 MB 4.0 GB 0 > > 9 1571.97 140.4 MB 6.0 GB 0 > > 10 3098.54 22.8 GB 26.9 GB 0 > > ``` > > > > > > > > > > > > > > > > > > > > > > > > I ran `cargo run --profile release-nonlto --bin mem_profile nlj` in https://github.com/ding-young/datafusion/tree/nlj-rewrite-tmp-review and adjusted v1/v2 with `static USE_NLJ_V2: bool = false;` > > Thanks for the profiler 👍🏼 I'll review that soon. > > I think Q4 and Q10 should not consume that much memory 😅 > > ```sql > -- q10: > SELECT * > FROM range(30000) AS t1 > FULL JOIN range(30000) AS t2 > ON (t1.value + t2.value) % 10 <> 0; > ``` > > It is supposed to only buffer (30k x 8Bytes), 20G RSS is not expected. let me double check the implementation... @ding-young For q10, the maxRSS I got using `/usr/bin/time -v` is lower than yours (16712272kb vs 44.7 GB). The code for `./target/release/join_limit_join_batch_size` should be identical to the "v1 (old)" code. ``` √ devhomeinsp ~/c/datafusion > /usr/bin/time -v ./target/release/join_limit_join_batch_size --maxrows 1 -c ' SELECT * FROM range(30000) AS t1 FULL JOIN range(30000) AS t2 ON (t1.value + t2.value) % 10 <> 0;' DataFusion CLI v48.0.0 +-------+-------+ | value | value | +-------+-------+ | 0 | 24576 | | . | | . | | . | +-------+-------+ 810000000 row(s) fetched. (First 1 displayed. Use --maxrows to adjust) Elapsed 4.202 seconds. Command being timed: "./target/release/join_limit_join_batch_size --maxrows 1 -c SELECT * FROM range(30000) AS t1 FULL JOIN range(30000) AS t2 ON (t1.value + t2.value) % 10 <> 0;" User time (seconds): 9.26 System time (seconds): 4.74 Percent of CPU this job got: 322% Elapsed (wall clock) time (h:mm:ss or m:ss): 0:04.34 Average shared text size (kbytes): 0 Average unshared data size (kbytes): 0 Average stack size (kbytes): 0 Average total size (kbytes): 0 Maximum resident set size (kbytes): 16712272 Average resident set size (kbytes): 0 Major (requiring I/O) page faults: 0 Minor (reclaiming a frame) page faults: 3028619 Voluntary context switches: 72488 Involuntary context switches: 28 Swaps: 0 File system inputs: 0 File system outputs: 0 Socket messages sent: 0 Socket messages received: 0 Signals delivered: 0 Page size (bytes): 4096 Exit status: 0 ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
